"Now let's imagine that gna.org were unavailable (or the project hosting) : how would I know it ? I cannot receive mail or see the gna! front page as they are unavailable..."

Well, Gna! is composed of several machines, they are unlikely to be all down at the same time. But even if it happens, admins of Gna! still have others machines, that cannot handle Gna! but that are enough to keep users informed.

The lack of information, in my experience, is more due to a state of mind than to technical problems.

"Otherwise you enforce slavery ;-)) which is an image of me being enslaved as I cannot be informed, cannot help, just have to wait... "

Hum, I have a different understanding of slavery, but I understand what you mean.

"becomes crucial once you have no way of communicating to your users (unless you have a backup plan with another server available that can take over and restore the service for example)."

We clearly do not have the money to have more than 3 machines dedicated to Gna!. So if the place where the computers are stored was burning, we would surely not be able to restablish the services the same day. However, this case is a bit extreme (it happened with the machine hosting security.debian.org once, if I remember well, but I cannot think of any other example) and we cannot afford an insurance against such unexpected troubles.

That's said, Gna! was designed and set up in a very short time, I guess that, if we had to do it one more time, I would be pretty fast.

So in case of "normal" trouble, like one machine providing services offline, we would move the services temporarily to a spare machine we have. In case of several machines offline, it depends on the cause of trouble.
We do not offer warrantee of a 24h/24h availability service, but we would clearly not let things pending for several weeks.
For instance, it was possible to get savannah.gnu.org restarting the 5th december, with limited but essential services like CVS and download areas. That the choice we would have made, considering that buying a new machine and making lot of hacks everywhere to make the whole stuff fit-in in a new architecture designed in a hurry was not the best way to go for user's sake.

When I was not involved in Gna! and more involved in Savannah savannah.gnu.org, no downtime for more than one day happened because I made choices to avoid that. I intend to make Gna! following that approach. Clearly, a downtime of one month is not an option in my mind. In fact, I dont understand how it can be an option to anyone, unless you think you are only dealing with amateurs (amateurs is not an insult, but amateurs expectations are obviously lower than professional ones, their salary is not at stakes -- but indeed, many amateurs out there are in fact professionals doing here a non-paid job, as amateurs).

Apart from that, projects should make backups of their data. We do store backup on several machines, but who knows?

I think we will in a short time propose a daily (or weekly) backup tarball (including sql and cvs data and arch data) that projects can keep, and that could be used to restore their project (even elsewhere).

Does it address your questions?